Mori, on your desktop.

The whole agent, on your Mac or Windows PC. She works your apps, your files and your real Chrome, builds with a live preview, and checks the result before she tells you it is done.

Studio Mori on a Mac after fixing a bug in a photo upload. The turn says how long it worked, what it found and what it changed, a card lists the three changed files, and the tasks panel on the right shows the run.
A bug she found and fixed: big photos failed to upload. The turn says how long it worked and what it changed, and the card lists every file it touched.

363

Built-in tools

Files, terminal, apps and windows, Chrome, git, documents and more.

189

Skills in the Hub

Ready procedures that install offline.

50

Slash commands

35 of them also work from Discord and Telegram.

16

Ready workflows

And four workflow scripts beside them.

29

Connectors

GitHub, Linear, Sentry, Notion, Figma, Blender and more.

09

Providers on your own key

Plus the managed plan and local Ollama.

12

Messaging channels

Discord, Telegram, Slack, Signal, email and more.

08

Helpers at once

In Swarm, merged into one report.

01  Your computer

She works your computer, the way you do.

She clicks, types and switches apps the way you would, and works inside the browser you are already signed into. To press a button she reads the window's own list of controls, then clicks the one she wants, with no screenshot needed.

Interactive elements of "TextEdit" (4):

  1. 1.[Button]Cancel
  2. 2.[Button]Save
  3. 3.[TextField]Save As
  4. 4.[PopUpButton]Where

click_element(2)

On the desktop

She drives the mouse, the keyboard, the windows and the apps, with the same tools on a Mac and on a Windows PC.

  • Buttons by name

    She lists the buttons, fields and menus of the front window from the system's accessibility tree and clicks them by name, with any model and no screenshot needed.

  • Every mouse and keyboard move

    Middle and triple clicks, drags built from a separate press and release, held keys and chords, and text fields set directly.

  • Apps and windows

    She lists running apps, opens, focuses and closes them, and reads or changes a window's position, size and state.

  • System controls

    Volume, mute, dark mode, sleeping the displays, locking the screen, and Wi-Fi on or off are one request away.

  • Screen awareness

    Each message can tell her which app and window you are in, your front Chrome tab and which terminals are open. Titles and the tab's address only, never a picture of your screen, and one switch turns it off.

  • Home and desk

    She controls Home Assistant devices, plays and skips music in Spotify or Apple Music, and adds Calendar events and Reminders on the Mac.

In your real Chrome

With the Mori Chrome extension, she works inside the Chrome you are already signed into.

  • Whole web tasks

    A web task runs in your signed-in Chrome with real clicks, typing and form fills. On a checkout she fills the cart and stops before the final payment.

  • Reads the real page

    The title, the headings, every interactive element including iframes, the links and the text of the page you actually have open.

  • Tabs

    She lists, opens, switches and closes your tabs, and can move a background tab to a new page without bringing it forward.

  • Console errors

    She reads console messages and JavaScript errors from a live page in your Chrome.

  • A separate browser for errands

    A separate, signed-out browser searches Amazon, eBay, Walmart, Best Buy, Target, Etsy and Google, compares listings, and stops at the cart so you finish any purchase.

  • The requests behind a page

    She can capture the JSON requests a page makes, to find where its data comes from or why it loads slowly.

02  Builds

Designs and builds, with a live preview.

Ask for a site, an app or a screen. She plans it like a designer, builds it, and looks at the result before she hands it over.

Studio Mori in the middle of a fix. A line counts the work so far, and each step has its own row: what Mori thought, explored, edited and created, and the test command still running.
Mid-turn. A line counts the work so far, and every step is a row you can open, down to the test that is still running.
  • Designer mode

    Say /design and she reads your project's own tokens, theme and components first, looks for references on 21st.dev, GitHub and the open web, and writes a short plan before she builds.

  • Honest references

    A reference source that could not be reached is named with its error, and no link appears that did not come back from a real fetch.

  • A designed starting point

    New web apps start from a Vite, React and Tailwind v4 project with design tokens, shadcn-style components, Lucide icons and Inter.

  • Live preview

    The Mori Preview window shows a project folder, an HTML file or a running dev server, and dev servers keep running in the background.

  • She looks at her own work

    She checks a build with a picture of the preview window only, never your desktop, and the picture comes with the page's console errors.

  • Creation

    Pictures, clips, sound, a timeline editor, 3D worlds, ads and workflows in one place, with every version kept and the price shown before anything is spent.

03  Code

Her own branch, tests that can fail, a pull request.

She reads a repo, works on her own branch, proves the tests can fail, and opens the pull request. The merge is yours.

#14 Keep the time zone in exported dates

mori/export-time-zone → main

CI passing (3)

Merge #14 into main?

Keep the time zone in exported dates
mori/export-time-zone → main. CI is green. This merges on GitHub and cannot be undone from here.

  • Knows a repo in one step

    One call reads a project's own instructions file, how it builds and tests, and its layout with file counts.

  • Workspaces

    A big or risky job runs in its own git worktree and branch, so your main checkout is untouched until you merge it back.

  • Pull requests

    She opens a pull request from a fresh mori/ branch, never commits on main, leaves merging to a button you confirm, and wakes the thread when CI fails so she can read the log and fix it.

  • Tests that can fail

    She breaks the code one change at a time and reruns the tests, so a change no test notices shows up as behaviour nothing checks.

  • Terminals and other agents

    Named shells persist across turns on your machine, in a throwaway Docker container or over SSH, and she can drive interactive programs like claude, codex, aider, Python or psql.

  • Plan first, review after

    /captain writes a spec in a read-only pass for you to approve before the build, and /review reads the folder's working diff and reports findings by severity.

04  Research

Research that reads its sources.

She reads the sources, not the snippets, and every answer ends with what it cited. A source it cites but never opened says so.

4 sources1 cited, not read

  • STRICT Tables

    sqlite.org

  • Release History Of SQLite

    sqlite.org

  • Datatypes In SQLite

    sqlite.org

  • github.com/sqlite/sqlite

    github.comcited, not read

  • Deep research

    She splits a question into angles, sends researchers out in parallel to search and read, and returns one answer with its sources.

  • Unread sources are marked

    A source that was cited but never read is marked, and a GitHub page read through GitHub's API counts as read.

  • Reads pages in full

    Up to five pages at once, as clean text, and long pages are kept whole rather than cut.

  • GitHub as a real source

    She searches repositories, searches real code once a GitHub token is set, reads files, reads issues including closed ones, and reads release notes.

  • Developer answers

    Packages on npm, PyPI and crates.io, the top-voted Stack Overflow answer, and Hacker News discussions.

  • Guesses are called guesses

    A date given beside a version that no page read in the run carries is flagged as unverified.

05  Helpers

Helpers for big jobs, workflows for repeat ones.

Big jobs split across helpers that each start with a clean context. Five modes say what kind of work a turn is, from Build to Swarm.

  • BuildThe default: researches, plans, builds and checks
  • SparkChat only, changes nothing
  • TranceResearch only: reads and cites, never writes
  • ForgeLong builds, with every tool loaded at once
  • SwarmA team of up to eight builds parts at once
  • Helpers

    An independent piece of a big job goes to a helper with its own fresh context and full tools, and comes back as a short result with file paths.

  • Swarm

    In Swarm, up to eight helpers work at once and their results merge into one report.

  • Four built-in agents

    Researcher reads and never writes, Coder builds and runs, Critic finds problems and never rewrites, and Designer shapes look and copy. You can write your own as markdown files.

  • Workflows, drawn live

    Each step of a workflow is drawn in the chat as it runs, and /workflow runs any ask as one: fan out, gather sources, verify, then report.

  • Workflow scripts

    Four script patterns ship ready: adversarial verify, judge panel, loop until dry, and sweep and critic. A script runs in a sandbox with no file access, network or clock, and asks before it starts.

  • Advisors

    /moa has other models draft first, at one extra call each per turn.

Find and fix a bug

  1. Pin down what actually goes wrong
  2. Get the lay of the project
  3. Find the code responsible
  4. Work out the real cause
  5. Make the smallest fix that addresses the cause
  6. Prove it is fixed

Sixteen workflows, ready to run.

  1. Find and fix a bug
  2. Understand an unfamiliar codebase
  3. Debug something that is failing
  4. Write tests for existing code
  5. Review and ship what you changed
  6. Find and remove dead code
  7. Check this machine is set up properly
  8. Answer a question properly, with sources
  9. See how something compares to its competitors
  10. Build a presentation
  11. Write up what happened this week
  12. Draft a message to someone, for you to approve
  13. Research the question
  14. Build it, then prove it works
  15. Review it properly
  16. Study a repo, then build on it
06  On its own

Work that keeps going while you are away.

Automations, routines and goals keep working while you are away, inside the limits you set.

  • Automations

    An automation is a saved prompt with one to twenty triggers: a schedule, on demand, a webhook, GitHub activity or a chat channel.

  • Six routines

    The routines here are ready to schedule with /blueprint, each at a time you can change.

  • Exact timing

    Jobs run on a five-field cron expression, once at a set moment, or every few minutes.

  • Writes only where you said

    When you schedule a job you can name the folders it may write in, and it writes nowhere else.

  • /goal

    Keeps a thread working until an independent check agrees the goal is met, and stops after six continuations, two dollars or thirty minutes.

  • Quiet unless something changed

    Scheduled jobs remember what they found last time, stay quiet when nothing changed, and retry when the model was unreachable.

  • Morning brief07:30 weekdaysA short read on your day before you start it.
  • Watch for important mailEvery 4 hoursTells you when something actually needs you.
  • Weekly reviewFridays 17:00What moved this week, and what quietly did not.
  • Watch for renewalsMondays 09:00A warning before a subscription charges.
  • Tidy the Downloads folderSaturdays 10:00Sorts the pile into folders, and deletes nothing.
  • Evening wind down18:30 weekdaysCloses the day out so tomorrow starts from somewhere.
07  Memory

She remembers, and brings the right know-how.

Facts, people and past work stay on your machine, and each ask brings in the skills written for it.

  • answers-first.md

    Wants the answer first, and detail only when asked

  • site-deploys.md

    The shop site deploys from main, and never on a Friday

  • brand-colours.md

    Deep green and warm white, with no gradients

  • Memory you can read

    Each fact is its own small file with a one-line recall key, and a near duplicate becomes an update instead of a second copy.

  • A brain with dates

    A knowledge graph links people, projects and things. She can walk its links, show an entity's dated timeline, or say when it was last seen as of any date.

  • Stays on your machine

    The brain ships inside the app and runs on your computer, with embeddings from local Ollama or a bundled local server unless you choose otherwise.

  • Every past conversation

    She searches the full text of every past conversation, so a passing mention from weeks ago is findable.

  • Skills come in on their own

    She reads what you asked for and brings in up to eight skills for it, from the 35 built-in ones or your own.

  • Skills Hub

    189 ready procedures ship with the app and install offline, and /learn turns a folder, a page, a topic or the last task into a new skill.

Thirty skills, built in.

  • Web project build
  • Deep research method
  • Polished deliverable
  • Debugging protocol
  • Ad campaign craft
  • Business outreach
  • Social content engine
  • Market and competitor scan
  • Blender scene by script
  • Game asset in Blender
  • Materials in Blender
  • Rig and animate in Blender
  • Blender render for a site or video
  • Design system discipline
  • Anti-slop rules
  • Design critique pass
  • Code project build
  • App build and verify
  • Chat bot build
  • Data analysis pass
  • Find what a build needs
  • Emulator and systems build
  • Video edit with Mori's editor
  • Buying automations
  • After Effects motion graphics by script
  • Premiere Pro edit and export by script
  • Photoshop by script
  • Adobe apps when scripting cannot reach
  • Commercial production
  • Campaign production
  • Interactive world building
  • Code and security review
  • Game mod build
  • Generated pictures and clips
  • Animate a still
08  Workspace

Projects, threads, and a look that is yours.

Work stays organised by project, and every thread shows what it is waiting on. The home screen shows what is running, what needs you, and your latest threads.

The Studio Mori home screen: projects and threads in the sidebar, a greeting over your own picture, one composer, a launchpad of tools, and the threads to continue.

/for commands

  • /designDesigner mode: real references and a plan first
  • /captainA read-only pass writes a spec you approve
  • /reviewReview the working diff, findings by severity
  • /goalKeep working until this holds
  • /projectDedicate this thread to a project
  • /learnTurn a folder, page or topic into a skill
  • /blueprintSet up a routine that runs on its own
  • /sessionsSearch the full text of every past chat
  • Projects

    A project has a name, a goal every chat in it inherits, and a folder where its work lands.

  • Dedicated and starred threads

    /project dedicates a thread so she knows its folder, repos and goal without being told again, and /star keeps a thread forever.

  • Threads that show their state

    Each thread shows whether it needs you, is working, is waiting, is ready for review or is idle, with a short code, up to twenty tags and its running cost.

  • More than one Mori

    Run up to nine copies, each with its own settings, memory, sign-in, ports and projects.

  • Your look

    Four layouts (simple, default, pro and focus) and ten themes, from Porcelain to ASCII, change the whole app.

  • Branch and export

    /branch forks a conversation, and /export saves it as Markdown or HTML with secrets scrubbed.

09  Files and voice

Reads almost any file. Makes real ones.

Hand her a file and she reads it. Ask for one and she writes a real file you can send. She can speak her replies, too.

Her voice, tried in order

  1. 1ElevenLabsOn your ElevenLabs key
  2. 2OpenAIOn your OpenAI key
  3. 3Local neural voiceWhen you have installed one
  4. 4Built-in system voiceAlways there, and always last
  • Documents, sheets and decks

    Word or PDF documents, Excel or CSV spreadsheets, and PowerPoint decks, written as real files.

  • Reads almost any file

    Code, images, PDFs, Word, PowerPoint, Excel, CSVs and zips are detected and read.

  • Data and scans

    Real SQL over a CSV, and the text in screenshots and scans, read with your system's own OCR.

  • Audio and video

    She transcribes voice notes and recordings up to 25 MB, and watches a video by sampling frames across the whole clip.

  • Voice

    Talk with an open mic or push to talk. Replies are spoken through ElevenLabs, OpenAI or a local voice, with your computer's own voice as the fallback.

  • Images, video and music

    Pictures, clips and music on your OpenRouter key, or on your own keys for Runway, Google, OpenAI, Black Forest Labs, fal, Replicate, Luma, Stability and ElevenLabs.

10  Everywhere

Reach her from anywhere.

Your phone, your chat apps and your other tools can all hand her work on this computer. About the iPhone app

Studio Mori on an iPhone: its Home, asking what you would like to build, with your paired Studio Mac, one box to ask Mori anything, and doors to build an app, research, make pictures, video and edits
  • Mori Link

    Pair your iPhone with a six-digit code while both are signed into the same account, and this computer runs commands only from that phone.

  • From your phone

    Once paired, the phone runs terminal commands, builds files, drives the browser, sends email and messages, and runs multi-step tasks with full memory and tools.

  • Discord and Telegram

    Up to eight Discord and Telegram bots, each with its own owner, and their menus carry 32 of the slash commands.

  • More channels

    Slack, Matrix, Signal, Zulip, Mattermost, IRC, ntfy, email, and iMessage on the Mac, plus any outbound webhook.

  • Email

    She sends email with attachments through the Gmail you are signed into in Chrome, with an SMTP account as the fallback, and sends in bulk up to 25 at a time.

  • For your other tools

    A local OpenAI-compatible API, an MCP server for Claude Code and editors, and an ACP mode for editors like Zed, each off until you switch it on.

11  Control

Pick the model. She does not stop to ask.

No approval cards: only your system asks, and her own rails refuse what should never run. Turn asking on for any chat that you want to approve step by step.

  • Don't askThe default. Edits, commands, installs and tests just run, on the record.
  • GuidedShows file changes as a diff first, and confirms the risky commands it recognises.
  • CommanderCommands run with no card, for a set time or with no limit.
  • Nine providers on your own key

    OpenRouter, Anthropic, OpenAI, Groq, DeepSeek, xAI, Together AI, Mistral or Perplexity on your own key, the managed plan included when you sign in, or local models through Ollama with no key.

  • The newest Claude

    On your own keys with the model on Auto, Quick is Sonnet 5.5 and Balanced, Deep and Max are Opus 5.5, thinking harder at each step. Fable 5.1 is there when you pick it.

  • No approval cards

    Edits, commands, installs and tests just run, and every step is on the record. Only your Mac or Windows asks, for its own permissions. Asking can be turned on for every chat or one.

  • Protected files

    Instruction files, git hooks, git settings, editor settings and CI workflows are changed only when your own words name them. A Safety rails switch turns her rails off, at your own risk.

  • Switches and undo

    Seven permission switches cover apps, browser, screen awareness, file writing, terminal, keyboard and destructive actions, and every file is checkpointed before she changes it, so the last few minutes can be undone.

  • Where the money goes

    A spend policy splits the work into building, research, chat, housekeeping, helpers and double-checking, with Saver, Balanced and Full presets.

12  Begin

Give her the whole job.